Fix: ust: UST communication can return -EAGAIN
[lttng-tools.git] / src / bin / lttng-sessiond / ust-consumer.c
1 /*
2 * Copyright (C) 2011 - David Goulet <david.goulet@polymtl.ca>
3 *
4 * This program is free software; you can redistribute it and/or modify
5 * it under the terms of the GNU General Public License, version 2 only,
6 * as published by the Free Software Foundation.
7 *
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
11 * GNU General Public License for more details.
12 *
13 * You should have received a copy of the GNU General Public License along
14 * with this program; if not, write to the Free Software Foundation, Inc.,
15 * 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
16 */
17
18 #define _LGPL_SOURCE
19 #include <errno.h>
20 #include <stdio.h>
21 #include <stdlib.h>
22 #include <string.h>
23 #include <unistd.h>
24 #include <inttypes.h>
25
26 #include <common/common.h>
27 #include <common/consumer/consumer.h>
28 #include <common/defaults.h>
29
30 #include "consumer.h"
31 #include "health-sessiond.h"
32 #include "ust-consumer.h"
33 #include "lttng-ust-error.h"
34 #include "buffer-registry.h"
35 #include "session.h"
36 #include "lttng-sessiond.h"
37
38 /*
39 * Send a single channel to the consumer using command ASK_CHANNEL_CREATION.
40 *
41 * Consumer socket lock MUST be acquired before calling this.
42 */
43 static int ask_channel_creation(struct ust_app_session *ua_sess,
44 struct ust_app_channel *ua_chan,
45 struct consumer_output *consumer,
46 struct consumer_socket *socket,
47 struct ust_registry_session *registry,
48 struct lttng_trace_chunk *trace_chunk)
49 {
50 int ret, output;
51 uint32_t chan_id;
52 uint64_t key, chan_reg_key;
53 char *pathname = NULL;
54 struct lttcomm_consumer_msg msg;
55 struct ust_registry_channel *chan_reg;
56 char shm_path[PATH_MAX] = "";
57 char root_shm_path[PATH_MAX] = "";
58 bool is_local_trace;
59
60 assert(ua_sess);
61 assert(ua_chan);
62 assert(socket);
63 assert(consumer);
64 assert(registry);
65
66 DBG2("Asking UST consumer for channel");
67
68 is_local_trace = consumer->net_seq_index == -1ULL;
69 /* Format the channel's path (relative to the current trace chunk). */
70 pathname = setup_channel_trace_path(consumer, ua_sess->path);
71 if (!pathname) {
72 ret = -1;
73 goto error;
74 }
75
76 if (is_local_trace && trace_chunk) {
77 enum lttng_trace_chunk_status chunk_status;
78 char *pathname_index;
79
80 ret = asprintf(&pathname_index, "%s/" DEFAULT_INDEX_DIR,
81 pathname);
82 if (ret < 0) {
83 ERR("Failed to format channel index directory");
84 ret = -1;
85 goto error;
86 }
87
88 /*
89 * Create the index subdirectory which will take care
90 * of implicitly creating the channel's path.
91 */
92 chunk_status = lttng_trace_chunk_create_subdirectory(
93 trace_chunk, pathname_index);
94 free(pathname_index);
95 if (chunk_status != LTTNG_TRACE_CHUNK_STATUS_OK) {
96 ret = -1;
97 goto error;
98 }
99 }
100
101 /* Depending on the buffer type, a different channel key is used. */
102 if (ua_sess->buffer_type == LTTNG_BUFFER_PER_UID) {
103 chan_reg_key = ua_chan->tracing_channel_id;
104 } else {
105 chan_reg_key = ua_chan->key;
106 }
107
108 if (ua_chan->attr.type == LTTNG_UST_CHAN_METADATA) {
109 chan_id = -1U;
110 /*
111 * Metadata channels shm_path (buffers) are handled within
112 * session daemon. Consumer daemon should not try to create
113 * those buffer files.
114 */
115 } else {
116 chan_reg = ust_registry_channel_find(registry, chan_reg_key);
117 assert(chan_reg);
118 chan_id = chan_reg->chan_id;
119 if (ua_sess->shm_path[0]) {
120 strncpy(shm_path, ua_sess->shm_path, sizeof(shm_path));
121 shm_path[sizeof(shm_path) - 1] = '\0';
122 strncat(shm_path, "/",
123 sizeof(shm_path) - strlen(shm_path) - 1);
124 strncat(shm_path, ua_chan->name,
125 sizeof(shm_path) - strlen(shm_path) - 1);
126 strncat(shm_path, "_",
127 sizeof(shm_path) - strlen(shm_path) - 1);
128 }
129 strncpy(root_shm_path, ua_sess->root_shm_path, sizeof(root_shm_path));
130 root_shm_path[sizeof(root_shm_path) - 1] = '\0';
131 }
132
133 switch (ua_chan->attr.output) {
134 case LTTNG_UST_MMAP:
135 default:
136 output = LTTNG_EVENT_MMAP;
137 break;
138 }
139
140 consumer_init_ask_channel_comm_msg(&msg,
141 ua_chan->attr.subbuf_size,
142 ua_chan->attr.num_subbuf,
143 ua_chan->attr.overwrite,
144 ua_chan->attr.switch_timer_interval,
145 ua_chan->attr.read_timer_interval,
146 ua_sess->live_timer_interval,
147 ua_sess->live_timer_interval != 0,
148 ua_chan->monitor_timer_interval,
149 output,
150 (int) ua_chan->attr.type,
151 ua_sess->tracing_id,
152 pathname,
153 ua_chan->name,
154 consumer->net_seq_index,
155 ua_chan->key,
156 registry->uuid,
157 chan_id,
158 ua_chan->tracefile_size,
159 ua_chan->tracefile_count,
160 ua_sess->id,
161 ua_sess->output_traces,
162 ua_sess->real_credentials.uid,
163 ua_chan->attr.blocking_timeout,
164 root_shm_path, shm_path,
165 trace_chunk,
166 &ua_sess->effective_credentials);
167
168 health_code_update();
169
170 ret = consumer_socket_send(socket, &msg, sizeof(msg));
171 if (ret < 0) {
172 goto error;
173 }
174
175 ret = consumer_recv_status_channel(socket, &key,
176 &ua_chan->expected_stream_count);
177 if (ret < 0) {
178 goto error;
179 }
180 /* Communication protocol error. */
181 assert(key == ua_chan->key);
182 /* We need at least one where 1 stream for 1 cpu. */
183 if (ua_sess->output_traces) {
184 assert(ua_chan->expected_stream_count > 0);
185 }
186
187 DBG2("UST ask channel %" PRIu64 " successfully done with %u stream(s)", key,
188 ua_chan->expected_stream_count);
189
190 error:
191 free(pathname);
192 health_code_update();
193 return ret;
194 }
195
196 /*
197 * Ask consumer to create a channel for a given session.
198 *
199 * Session list and rcu read side locks must be held by the caller.
200 *
201 * Returns 0 on success else a negative value.
202 */
203 int ust_consumer_ask_channel(struct ust_app_session *ua_sess,
204 struct ust_app_channel *ua_chan,
205 struct consumer_output *consumer,
206 struct consumer_socket *socket,
207 struct ust_registry_session *registry,
208 struct lttng_trace_chunk * trace_chunk)
209 {
210 int ret;
211
212 assert(ua_sess);
213 assert(ua_chan);
214 assert(consumer);
215 assert(socket);
216 assert(registry);
217
218 if (!consumer->enabled) {
219 ret = -LTTNG_ERR_NO_CONSUMER;
220 DBG3("Consumer is disabled");
221 goto error;
222 }
223
224 pthread_mutex_lock(socket->lock);
225 ret = ask_channel_creation(ua_sess, ua_chan, consumer, socket, registry,
226 trace_chunk);
227 pthread_mutex_unlock(socket->lock);
228 if (ret < 0) {
229 ERR("ask_channel_creation consumer command failed");
230 goto error;
231 }
232
233 error:
234 return ret;
235 }
236
237 /*
238 * Send a get channel command to consumer using the given channel key. The
239 * channel object is populated and the stream list.
240 *
241 * Return 0 on success else a negative value.
242 */
243 int ust_consumer_get_channel(struct consumer_socket *socket,
244 struct ust_app_channel *ua_chan)
245 {
246 int ret;
247 struct lttcomm_consumer_msg msg;
248
249 assert(ua_chan);
250 assert(socket);
251
252 memset(&msg, 0, sizeof(msg));
253 msg.cmd_type = LTTNG_CONSUMER_GET_CHANNEL;
254 msg.u.get_channel.key = ua_chan->key;
255
256 pthread_mutex_lock(socket->lock);
257 health_code_update();
258
259 /* Send command and wait for OK reply. */
260 ret = consumer_send_msg(socket, &msg);
261 if (ret < 0) {
262 goto error;
263 }
264
265 /* First, get the channel from consumer. */
266 ret = ustctl_recv_channel_from_consumer(*socket->fd_ptr, &ua_chan->obj);
267 if (ret < 0) {
268 if (ret != -EPIPE) {
269 ERR("Error recv channel from consumer %d with ret %d",
270 *socket->fd_ptr, ret);
271 } else {
272 DBG3("UST app recv channel from consumer. Consumer is dead.");
273 }
274 goto error;
275 }
276
277 /* Next, get all streams. */
278 while (1) {
279 struct ust_app_stream *stream;
280
281 /* Create UST stream */
282 stream = ust_app_alloc_stream();
283 if (stream == NULL) {
284 ret = -ENOMEM;
285 goto error;
286 }
287
288 /* Stream object is populated by this call if successful. */
289 ret = ustctl_recv_stream_from_consumer(*socket->fd_ptr, &stream->obj);
290 if (ret < 0) {
291 free(stream);
292 if (ret == -LTTNG_UST_ERR_NOENT) {
293 DBG3("UST app consumer has no more stream available");
294 break;
295 }
296 if (ret != -EPIPE) {
297 ERR("Recv stream from consumer %d with ret %d",
298 *socket->fd_ptr, ret);
299 } else {
300 DBG3("UST app recv stream from consumer. Consumer is dead.");
301 }
302 goto error;
303 }
304
305 /* Order is important this is why a list is used. */
306 cds_list_add_tail(&stream->list, &ua_chan->streams.head);
307 ua_chan->streams.count++;
308
309 DBG2("UST app stream %d received successfully", ua_chan->streams.count);
310 }
311
312 /* This MUST match or else we have a synchronization problem. */
313 assert(ua_chan->expected_stream_count == ua_chan->streams.count);
314
315 /* Wait for confirmation that we can proceed with the streams. */
316 ret = consumer_recv_status_reply(socket);
317 if (ret < 0) {
318 goto error;
319 }
320
321 error:
322 health_code_update();
323 pthread_mutex_unlock(socket->lock);
324 return ret;
325 }
326
327 /*
328 * Send a destroy channel command to consumer using the given channel key.
329 *
330 * Note that this command MUST be used prior to a successful
331 * LTTNG_CONSUMER_GET_CHANNEL because once this command is done successfully,
332 * the streams are dispatched to the consumer threads and MUST be teardown
333 * through the hang up process.
334 *
335 * Return 0 on success else a negative value.
336 */
337 int ust_consumer_destroy_channel(struct consumer_socket *socket,
338 struct ust_app_channel *ua_chan)
339 {
340 int ret;
341 struct lttcomm_consumer_msg msg;
342
343 assert(ua_chan);
344 assert(socket);
345
346 memset(&msg, 0, sizeof(msg));
347 msg.cmd_type = LTTNG_CONSUMER_DESTROY_CHANNEL;
348 msg.u.destroy_channel.key = ua_chan->key;
349
350 pthread_mutex_lock(socket->lock);
351 health_code_update();
352
353 ret = consumer_send_msg(socket, &msg);
354 if (ret < 0) {
355 goto error;
356 }
357
358 error:
359 health_code_update();
360 pthread_mutex_unlock(socket->lock);
361 return ret;
362 }
363
364 /*
365 * Send a given stream to UST tracer.
366 *
367 * On success return 0 else a negative value.
368 */
369 int ust_consumer_send_stream_to_ust(struct ust_app *app,
370 struct ust_app_channel *channel, struct ust_app_stream *stream)
371 {
372 int ret;
373
374 assert(app);
375 assert(stream);
376 assert(channel);
377
378 DBG2("UST consumer send stream to app %d", app->sock);
379
380 /* Relay stream to application. */
381 pthread_mutex_lock(&app->sock_lock);
382 ret = ustctl_send_stream_to_ust(app->sock, channel->obj, stream->obj);
383 pthread_mutex_unlock(&app->sock_lock);
384 if (ret < 0) {
385 if (ret == -EPIPE || ret == -LTTNG_UST_ERR_EXITING) {
386 DBG3("UST app send stream to ust failed. Application is dead. (pid: %d, sock: %d).",
387 app->pid, app->sock);
388 } else if (ret == -EAGAIN) {
389 WARN("UST app send stream to ust failed. Communication time out (pid: %d, sock: %d).",
390 app->pid, app->sock);
391 } else {
392 ERR("UST app send stream, handle %d, to ust failed with ret %d (pid: %d, sock: %d).",
393 stream->obj->handle, ret, app->pid,
394 app->sock);
395 }
396 goto error;
397 }
398 channel->handle = channel->obj->handle;
399
400 error:
401 return ret;
402 }
403
404 /*
405 * Send channel previously received from the consumer to the UST tracer.
406 *
407 * On success return 0 else a negative value.
408 */
409 int ust_consumer_send_channel_to_ust(struct ust_app *app,
410 struct ust_app_session *ua_sess, struct ust_app_channel *channel)
411 {
412 int ret;
413
414 assert(app);
415 assert(ua_sess);
416 assert(channel);
417 assert(channel->obj);
418
419 DBG2("UST app send channel to sock %d pid %d (name: %s, key: %" PRIu64 ")",
420 app->sock, app->pid, channel->name, channel->tracing_channel_id);
421
422 /* Send stream to application. */
423 pthread_mutex_lock(&app->sock_lock);
424 ret = ustctl_send_channel_to_ust(app->sock, ua_sess->handle, channel->obj);
425 pthread_mutex_unlock(&app->sock_lock);
426 if (ret < 0) {
427 if (ret == -EPIPE || ret == -LTTNG_UST_ERR_EXITING) {
428 DBG3("UST app send channel to ust failed. Application is dead (pid: %d, sock: %d).",
429 app->pid, app->sock);
430 } else if (ret == -EAGAIN) {
431 WARN("UST app send channel to ust failed. Communication timeout (pid: %d, sock: %d).",
432 app->pid, app->sock);
433 } else {
434 ERR("UST app send channel %s, to ust failed with ret %d (pid: %d, sock: %d).",
435 channel->name, ret, app->pid,
436 app->sock);
437 }
438 goto error;
439 }
440
441 error:
442 return ret;
443 }
444
445 /*
446 * Handle the metadata requests from the UST consumer
447 *
448 * Return 0 on success else a negative value.
449 */
450 int ust_consumer_metadata_request(struct consumer_socket *socket)
451 {
452 int ret;
453 ssize_t ret_push;
454 struct lttcomm_metadata_request_msg request;
455 struct buffer_reg_uid *reg_uid;
456 struct ust_registry_session *ust_reg;
457 struct lttcomm_consumer_msg msg;
458
459 assert(socket);
460
461 rcu_read_lock();
462 health_code_update();
463
464 /* Wait for a metadata request */
465 pthread_mutex_lock(socket->lock);
466 ret = consumer_socket_recv(socket, &request, sizeof(request));
467 pthread_mutex_unlock(socket->lock);
468 if (ret < 0) {
469 goto end;
470 }
471
472 DBG("Metadata request received for session %" PRIu64 ", key %" PRIu64,
473 request.session_id, request.key);
474
475 reg_uid = buffer_reg_uid_find(request.session_id,
476 request.bits_per_long, request.uid);
477 if (reg_uid) {
478 ust_reg = reg_uid->registry->reg.ust;
479 } else {
480 struct buffer_reg_pid *reg_pid =
481 buffer_reg_pid_find(request.session_id_per_pid);
482 if (!reg_pid) {
483 DBG("PID registry not found for session id %" PRIu64,
484 request.session_id_per_pid);
485
486 memset(&msg, 0, sizeof(msg));
487 msg.cmd_type = LTTNG_ERR_UND;
488 pthread_mutex_lock(socket->lock);
489 (void) consumer_send_msg(socket, &msg);
490 pthread_mutex_unlock(socket->lock);
491 /*
492 * This is possible since the session might have been destroyed
493 * during a consumer metadata request. So here, return gracefully
494 * because the destroy session will push the remaining metadata to
495 * the consumer.
496 */
497 ret = 0;
498 goto end;
499 }
500 ust_reg = reg_pid->registry->reg.ust;
501 }
502 assert(ust_reg);
503
504 pthread_mutex_lock(&ust_reg->lock);
505 ret_push = ust_app_push_metadata(ust_reg, socket, 1);
506 pthread_mutex_unlock(&ust_reg->lock);
507 if (ret_push == -EPIPE) {
508 DBG("Application or relay closed while pushing metadata");
509 } else if (ret_push < 0) {
510 ERR("Pushing metadata");
511 ret = -1;
512 goto end;
513 } else {
514 DBG("UST Consumer metadata pushed successfully");
515 }
516 ret = 0;
517
518 end:
519 rcu_read_unlock();
520 return ret;
521 }
This page took 0.03976 seconds and 4 git commands to generate.